Contents

2. Deep Learning: A Simple Example

  • Let’s get back to the Name Gender Classifier.

2.1. Prepare Data

## Packages Dependencies
import os
import shutil
import numpy as np
import nltk
from nltk.corpus import names
import random

from sklearn.model_selection import train_test_split
from sklearn.manifold import TSNE

import seaborn as sns
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib

matplotlib.rcParams['figure.dpi'] = 150
    
from lime.lime_text import LimeTextExplainer

import tensorflow as tf
import tensorflow.keras as keras

from keras.preprocessing.text import Tokenizer
from keras.preprocessing import sequence
from keras.utils import to_categorical, plot_model
from keras.models import Sequential
from keras import layers
    # from keras.layers import Dense
    # from keras.layers import LSTM, RNN, GRU
    # from keras.layers import Embedding
    # from keras.layers import SpatialDropout1D

import kerastuner

labeled_names = ([(name, 1) for name in names.words('male.txt')] +
                 [(name, 0) for name in names.words('female.txt')])
random.shuffle(labeled_names)

2.2. Train-Test Split

train_set, test_set = train_test_split(labeled_names,
                                       test_size=0.2,
                                       random_state=42)
print(len(train_set), len(test_set))

6355 1589

names = [n for (n, l) in train_set] ## X
labels = [l for (n, l) in train_set] ## y

len(names)

6355

2.3. Tokenizer

  • keras.preprocessing.text.Tokenizer is a very useful tokenizer for text processing in deep learning.

  • Tokenizer assumes that the word tokens of the input texts have been delimited by whitespaces.

  • Tokenizer provides the following functions:

    • It will first create a dictionary for the entire corpus (a mapping of each word token and its unique integer index index) (Tokenizer.fit_on_text())

    • It can then use the corpus dictionary to convert words in each corpus text into integer sequences (Tokenizer.texts_to_sequences())

    • The dictionary is in Tokenizer.word_index

  • Notes on Tokenizer:

    • By default, the token index 0 is reserved for padding token.

    • If oov_token is specified, it is default to index 1. (Default: oov_token=False)

    • Specify num_words=N for Tokenizer to include only top N words in converting texts to sequences.

    • Tokenizer will automatically remove punctuations.

    • Tokenizer use the whitespace as word-token delimiter.

    • If every character is treated as a token, specify char_level=True.

    • Please read Tokenizer documentation very carefully. Very important!!

tokenizer = Tokenizer(char_level=True)
tokenizer.fit_on_texts(names) ## similar to CountVectorizer.fit_transform()

Vocabulary

After we Tokenizer.fit_on_texts(), we can take a look at the corpus dictionary, i.e., the mapping of words and their unique integer indices.

# determine the vocabulary size
vocab_size = len(tokenizer.word_index) + 1
print('Vocabulary Size: %d' % vocab_size)

Vocabulary Size: 30

tokenizer.word_index

{'a': 1,
 'e': 2,
 'i': 3,
 'n': 4,
 'r': 5,
 'l': 6,
 'o': 7,
 't': 8,
 's': 9,
 'd': 10,
 'y': 11,
 'm': 12,
 'h': 13,
 'c': 14,
 'b': 15,
 'u': 16,
 'g': 17,
 'k': 18,
 'j': 19,
 'v': 20,
 'f': 21,
 'p': 22,
 'w': 23,
 'z': 24,
 'x': 25,
 'q': 26,
 '-': 27,
 ' ': 28,
 "'": 29}

2.4. Prepare Input and Output Tensors

  • Like in feature-based machine translation, a computational model only accepts numeric values. It is necessary to convert raw texts to numeric tensors for neural network.

  • After we create the Tokenizer, we use the Tokenizer to perform text vectorization, i.e., converting texts into tensors.

  • In deep learning, words or characters are automatically converted into numeric representations. In other words, the feature engineering step is fully automatic.

Two Ways of Text Vectorization

  • Texts to Sequences:

    • For each text, we convert all word tokens into integer sequences.

    • These integer sequences will then be transformed into embeddings in the deep learning network.

    • These embeddings are usually the basis for deep learning sequence models (i.e., RNN).

  • Texts to Matrix:

    • For each text, we vectorize the entire text into a vector of bag-of-words representation.

    • A One-hot encoding of the entire text would have all values of the text vector to be either 0 or 1, indicating the occurrence of all the words in the dictionary.

    • We can of course use the frequency-based bag-of-words representation (similar to CountVectorizer()).

2.5. Method 1: Text to Sequences

From Texts and Sequences

  • We can convert our corpus texts into integer sequences using Tokenizer.texts_to_sequences().

  • Because texts vary in lengths, we use keras.preprocessing.sequence.pad_sequence() to pad all texts into a uniform length.

  • This step is important. This ensures that every text, when pushed into the network, has exactly the same tensor shape.

names_ints = tokenizer.texts_to_sequences(names)

print(names[:5])
print(names_ints[:5])
print(labels[:5])

['Chanda', 'Sibley', 'Barbabra', 'Sherwin', 'Ginni']
[[14, 13, 1, 4, 10, 1], [9, 3, 15, 6, 2, 11], [15, 1, 5, 15, 1, 15, 5, 1], [9, 13, 2, 5, 23, 3, 4], [17, 3, 4, 4, 3]]
[0, 0, 0, 1, 0]

Padding

  • When padding the all texts into uniform lengths, consider whether to Pre-padding or removing values from the beginning of the sequence (i.e., pre) or the other way (post).

  • Check padding and truncating parameters in pad_sequences

  • In this tutorial, we first identify the longest name, and use its length as the max_len and pad all names into the max_len.

## We can check the length distribution of texts in corpus

names_lens = [len(n) for n in names_ints]
names_lens

sns.displot(names_lens)
print(names[np.argmax(names_lens)])  # longest name

Helen-Elizabeth
../_images/dl-simple-case_25_1.png 
max_len = names_lens[np.argmax(names_lens)]
max_len

15

names_ints_pad = sequence.pad_sequences(names_ints, maxlen=max_len)
names_ints_pad[:10]

array([[ 0,  0,  0,  0,  0,  0,  0,  0,  0, 14, 13,  1,  4, 10,  1],
       [ 0,  0,  0,  0,  0,  0,  0,  0,  0,  9,  3, 15,  6,  2, 11],
       [ 0,  0,  0,  0,  0,  0,  0, 15,  1,  5, 15,  1, 15,  5,  1],
       [ 0,  0,  0,  0,  0,  0,  0,  0,  9, 13,  2,  5, 23,  3,  4],
       [ 0,  0,  0,  0,  0,  0,  0,  0,  0,  0, 17,  3,  4,  4,  3],
       [ 0,  0,  0,  0,  0,  0,  0,  0, 23,  3,  6, 15,  2,  5,  8],
       [ 0,  0,  0,  0,  0,  0,  0,  0,  0, 22,  1, 12, 12,  3,  2],
       [ 0,  0,  0,  0,  0,  0,  0,  0,  0,  8,  7,  5,  5,  2, 11],
       [ 0,  0,  0,  0,  0,  0,  0, 14, 13,  5,  3,  9,  8,  3,  2],
       [ 0,  0,  0,  0,  0,  0,  0,  0,  0,  6,  1, 23,  8,  7,  4]],
      dtype=int32)

Define X and Y

  • So for names, we convert names into integer sequences, and pad them into the uniform length.

  • We perform exactly the same processing to the names in test data.

  • We convert both the names (X) and labels (y) into numpy.array.

## training data
X_train = np.array(names_ints_pad).astype('int32')
y_train = np.array(labels)


## testing data
X_test_texts = [n for (n, l) in test_set]
X_test = np.array(
    sequence.pad_sequences(tokenizer.texts_to_sequences(X_test_texts),
                           maxlen=max_len)).astype('int32')
y_test = np.array([l for (n, l) in test_set])

print(X_train.shape)
print(y_train.shape)
print(X_test.shape)
print(y_test.shape)

(6355, 15)
(6355,)
(1589, 15)
(1589,)

2.6. Method 2: Text to Matrix

One-Hot Encoding (Bag-of-Words)

  • We can convert each text to a bag-of-words vector using Tokenzier.texts_to_matrix().

  • In particular, we can specify the parameter mode: binary, count, or tfidf.

  • When the mode="binary", the text vector is a one-hot encoding vector, indicating whether a character occurs in the text or not.

names_matrix = tokenizer.texts_to_matrix(names, mode="binary")
print(names_matrix.shape)

(6355, 30)

print(names[2])
print(names_matrix[2,:])

Barbabra
[0. 1. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0.]

tokenizer.word_index

{'a': 1,
 'e': 2,
 'i': 3,
 'n': 4,
 'r': 5,
 'l': 6,
 'o': 7,
 't': 8,
 's': 9,
 'd': 10,
 'y': 11,
 'm': 12,
 'h': 13,
 'c': 14,
 'b': 15,
 'u': 16,
 'g': 17,
 'k': 18,
 'j': 19,
 'v': 20,
 'f': 21,
 'p': 22,
 'w': 23,
 'z': 24,
 'x': 25,
 'q': 26,
 '-': 27,
 ' ': 28,
 "'": 29}

Define X and Y

X_train2 = np.array(names_matrix).astype('int32')
y_train2 = np.array(labels)

X_test2 = tokenizer.texts_to_matrix(X_test_texts,
                                    mode="binary").astype('int32')
y_test2 = np.array([l for (n, l) in test_set])

print(X_train2.shape)
print(y_train2.shape)
print(X_test2.shape)
print(y_test2.shape)

(6355, 30)
(6355,)
(1589, 30)
(1589,)

2.7. Model Definition

  • Three important steps for building a deep neural network:

    • Define the model structure

    • Compile the model

    • Fit the model

  • After we have defined our input and output tensors (X and y), we can define the architecture of our neural network model.

  • For the two ways of name vectorized representations, we try two different types of networks.

    • Text to Matrix: Fully connected Dense Layers

    • Text to Sequences: Embedding + RNN

# Two Versions of Plotting Functions for `history` from `model.fit()`
def plot1(history):
    acc = history.history['accuracy']
    val_acc = history.history['val_accuracy']
    loss = history.history['loss']
    val_loss = history.history['val_loss']

    epochs = range(1, len(acc) + 1)
    ## Accuracy plot
    plt.plot(epochs, acc, 'bo', label='Training acc')
    plt.plot(epochs, val_acc, 'b', label='Validation acc')
    plt.title('Training and validation accuracy')
    plt.legend()
    ## Loss plot
    plt.figure()

    plt.plot(epochs, loss, 'bo', label='Training loss')
    plt.plot(epochs, val_loss, 'b', label='Validation loss')
    plt.title('Training and validation loss')
    plt.legend()
    plt.show()


def plot2(history):
    pd.DataFrame(history.history).plot(figsize=(8, 5))
    plt.grid(True)
    #plt.gca().set_ylim(0,1)
    plt.show()

Model 1: Fully Connected Dense Layers

  • Let’s try a simple neural network with two fully-connected dense layers with the Text-to-Matrix inputs.

  • That is, the input of this model is the bag-of-words representation of the entire name.

Dense Layer Operation

  • The transformation of each Dense layer will transform the input tensor into a tensor whose dimension size is the same as the node number of the Dense layer.

## Define Model
model1 = keras.Sequential()
model1.add(keras.Input(shape=(vocab_size, ), name="one_hot_input"))
model1.add(layers.Dense(16, activation="relu", name="dense_layer_1"))
model1.add(layers.Dense(16, activation="relu", name="dense_layer_2"))
model1.add(layers.Dense(1, activation="sigmoid", name="output"))

## Compile Model
model1.compile(loss=keras.losses.BinaryCrossentropy(),
               optimizer=keras.optimizers.Adam(lr=0.001),
               metrics=["accuracy"])

plot_model(model1, show_shapes=True)
../_images/dl-simple-case_49_0.png

A few hyperparameters for network training

  • Batch Size: The number of inputs needed per update of the model parameter (gradient descent)

  • Epoch: How many iterations needed for training

  • Validation Split Ratio: Proportion of validation and training data split

## Hyperparameters
BATCH_SIZE = 128
EPOCHS = 20
VALIDATION_SPLIT = 0.2

## Fit the model
history1 = model1.fit(X_train2,
                      y_train2,
                      batch_size=BATCH_SIZE,
                      epochs=EPOCHS,
                      verbose=2,
                      validation_split=VALIDATION_SPLIT)

Epoch 1/20
40/40 - 2s - loss: 0.6717 - accuracy: 0.6279 - val_loss: 0.6774 - val_accuracy: 0.6121
Epoch 2/20
40/40 - 0s - loss: 0.6582 - accuracy: 0.6279 - val_loss: 0.6676 - val_accuracy: 0.6121
Epoch 3/20
40/40 - 0s - loss: 0.6468 - accuracy: 0.6298 - val_loss: 0.6569 - val_accuracy: 0.6145
Epoch 4/20
40/40 - 0s - loss: 0.6334 - accuracy: 0.6379 - val_loss: 0.6426 - val_accuracy: 0.6271
Epoch 5/20
40/40 - 0s - loss: 0.6176 - accuracy: 0.6621 - val_loss: 0.6256 - val_accuracy: 0.6475
Epoch 6/20
40/40 - 0s - loss: 0.6018 - accuracy: 0.6817 - val_loss: 0.6135 - val_accuracy: 0.6648
Epoch 7/20
40/40 - 0s - loss: 0.5914 - accuracy: 0.6937 - val_loss: 0.6061 - val_accuracy: 0.6743
Epoch 8/20
40/40 - 0s - loss: 0.5827 - accuracy: 0.6996 - val_loss: 0.6012 - val_accuracy: 0.6814
Epoch 9/20
40/40 - 0s - loss: 0.5767 - accuracy: 0.7022 - val_loss: 0.5976 - val_accuracy: 0.6924
Epoch 10/20
40/40 - 0s - loss: 0.5727 - accuracy: 0.7042 - val_loss: 0.5948 - val_accuracy: 0.6955
Epoch 11/20
40/40 - 0s - loss: 0.5687 - accuracy: 0.7059 - val_loss: 0.5937 - val_accuracy: 0.6947
Epoch 12/20
40/40 - 0s - loss: 0.5658 - accuracy: 0.7105 - val_loss: 0.5935 - val_accuracy: 0.6939
Epoch 13/20
40/40 - 0s - loss: 0.5639 - accuracy: 0.7150 - val_loss: 0.5928 - val_accuracy: 0.6939
Epoch 14/20
40/40 - 0s - loss: 0.5619 - accuracy: 0.7146 - val_loss: 0.5932 - val_accuracy: 0.6924
Epoch 15/20
40/40 - 0s - loss: 0.5599 - accuracy: 0.7166 - val_loss: 0.5919 - val_accuracy: 0.6994
Epoch 16/20
40/40 - 0s - loss: 0.5577 - accuracy: 0.7175 - val_loss: 0.5915 - val_accuracy: 0.6971
Epoch 17/20
40/40 - 0s - loss: 0.5560 - accuracy: 0.7205 - val_loss: 0.5924 - val_accuracy: 0.6924
Epoch 18/20
40/40 - 0s - loss: 0.5545 - accuracy: 0.7213 - val_loss: 0.5924 - val_accuracy: 0.6924
Epoch 19/20
40/40 - 0s - loss: 0.5532 - accuracy: 0.7234 - val_loss: 0.5919 - val_accuracy: 0.6932
Epoch 20/20
40/40 - 0s - loss: 0.5521 - accuracy: 0.7215 - val_loss: 0.5919 - val_accuracy: 0.6947

plot1(history1)
../_images/dl-simple-case_53_0.png ../_images/dl-simple-case_53_1.png 
model1.evaluate(X_test2, y_test2, batch_size=128, verbose=2)

13/13 - 0s - loss: 0.5515 - accuracy: 0.7181

[0.5515258312225342, 0.7180616855621338]

Model 2: Embedding + RNN

  • Another possibility is to introduce an embedding layer in the network, which transforms each character in the name into a tensor (i.e., embeddings), and then to add a Recurrent Neural Network layer to process each character sequentially.

  • The strength of the RNN is that it iterates over the timesteps of a sequence, while maintaining an internal state that encodes information about the timesteps it has seen so far.

  • It is posited that after the RNN iterates through the entire sequence, it keeps important information of all previously iterated tokens for further operation.

  • The input of this network is a padded sequence of the original text (name).

## Define the embedding dimension
EMBEDDING_DIM = 128

## Define model
model2 = Sequential()
model2.add(
    layers.Embedding(input_dim=vocab_size,
                     output_dim=EMBEDDING_DIM,
                     input_length=max_len,
                     mask_zero=True))
model2.add(layers.SimpleRNN(16, activation="relu", name="RNN_layer"))
model2.add(layers.Dense(16, activation="relu", name="dense_layer"))
model2.add(layers.Dense(1, activation="sigmoid", name="output"))

model2.compile(loss=keras.losses.BinaryCrossentropy(),
               optimizer=keras.optimizers.Adam(lr=0.001),
               metrics=["accuracy"])

Embedding Layer Operation

RNN Layer Operation

RNN Layer Operation

Unrolled Version of RNN Operation

Unrolled Version of RNN Operation

plot_model(model2, show_shapes=True)
../_images/dl-simple-case_64_0.png 
history2 = model2.fit(X_train,
                      y_train,
                      batch_size=BATCH_SIZE,
                      epochs=EPOCHS,
                      verbose=2,
                      validation_split=VALIDATION_SPLIT)

Epoch 1/20
40/40 - 2s - loss: 0.6078 - accuracy: 0.6898 - val_loss: 0.5463 - val_accuracy: 0.7089
Epoch 2/20
40/40 - 0s - loss: 0.4753 - accuracy: 0.7707 - val_loss: 0.4726 - val_accuracy: 0.7569
Epoch 3/20
40/40 - 0s - loss: 0.4303 - accuracy: 0.7886 - val_loss: 0.4529 - val_accuracy: 0.7836
Epoch 4/20
40/40 - 0s - loss: 0.4119 - accuracy: 0.7994 - val_loss: 0.4458 - val_accuracy: 0.7884
Epoch 5/20
40/40 - 0s - loss: 0.4021 - accuracy: 0.8118 - val_loss: 0.4410 - val_accuracy: 0.7860
Epoch 6/20
40/40 - 0s - loss: 0.3963 - accuracy: 0.8141 - val_loss: 0.4386 - val_accuracy: 0.7852
Epoch 7/20
40/40 - 0s - loss: 0.3859 - accuracy: 0.8161 - val_loss: 0.4385 - val_accuracy: 0.7884
Epoch 8/20
40/40 - 0s - loss: 0.3813 - accuracy: 0.8224 - val_loss: 0.4352 - val_accuracy: 0.7923
Epoch 9/20
40/40 - 0s - loss: 0.3771 - accuracy: 0.8245 - val_loss: 0.4342 - val_accuracy: 0.7923
Epoch 10/20
40/40 - 0s - loss: 0.3728 - accuracy: 0.8253 - val_loss: 0.4339 - val_accuracy: 0.7907
Epoch 11/20
40/40 - 0s - loss: 0.3688 - accuracy: 0.8301 - val_loss: 0.4328 - val_accuracy: 0.7946
Epoch 12/20
40/40 - 0s - loss: 0.3659 - accuracy: 0.8283 - val_loss: 0.4302 - val_accuracy: 0.7915
Epoch 13/20
40/40 - 0s - loss: 0.3653 - accuracy: 0.8314 - val_loss: 0.4302 - val_accuracy: 0.7986
Epoch 14/20
40/40 - 0s - loss: 0.3579 - accuracy: 0.8322 - val_loss: 0.4318 - val_accuracy: 0.7915
Epoch 15/20
40/40 - 0s - loss: 0.3556 - accuracy: 0.8391 - val_loss: 0.4309 - val_accuracy: 0.7962
Epoch 16/20
40/40 - 0s - loss: 0.3534 - accuracy: 0.8375 - val_loss: 0.4298 - val_accuracy: 0.7954
Epoch 17/20
40/40 - 0s - loss: 0.3492 - accuracy: 0.8391 - val_loss: 0.4304 - val_accuracy: 0.7978
Epoch 18/20
40/40 - 0s - loss: 0.3466 - accuracy: 0.8413 - val_loss: 0.4322 - val_accuracy: 0.7946
Epoch 19/20
40/40 - 0s - loss: 0.3445 - accuracy: 0.8421 - val_loss: 0.4291 - val_accuracy: 0.7962
Epoch 20/20
40/40 - 0s - loss: 0.3413 - accuracy: 0.8421 - val_loss: 0.4286 - val_accuracy: 0.7986

plot1(history2)
../_images/dl-simple-case_66_0.png ../_images/dl-simple-case_66_1.png 
model2.evaluate(X_test, y_test, batch_size=128, verbose=2)

13/13 - 0s - loss: 0.4266 - accuracy: 0.7911

[0.4266381561756134, 0.7910635471343994]

Model 3: Regularization and Dropout

  • Based on the validation results of the previous two models (esp. the RNN-based model), we can see that they are probably a bit overfit because the model performance on the validation set starts to stall after the first few epochs.

  • We can add regularization and dropouts in our network definition to avoid overfitting.

## Define embedding dimension
EMBEDDING_DIM = 128

## Define model
model3 = Sequential()
model3.add(
    layers.Embedding(input_dim=vocab_size,
                     output_dim=EMBEDDING_DIM,
                     input_length=max_len,
                     mask_zero=True))
model3.add(
    layers.SimpleRNN(16,
                     activation="relu",
                     name="RNN_layer",
                     dropout=0.2, ## dropout for input character
                     recurrent_dropout=0.2))  ## dropout for previous state
model3.add(layers.Dense(16, activation="relu", name="dense_layer"))
model3.add(layers.Dense(1, activation="sigmoid", name="output"))

model3.compile(loss=keras.losses.BinaryCrossentropy(),
               optimizer=keras.optimizers.Adam(lr=0.001),
               metrics=["accuracy"])

plot_model(model3)
../_images/dl-simple-case_71_0.png 
history3 = model3.fit(X_train,
                      y_train,
                      batch_size=BATCH_SIZE,
                      epochs=EPOCHS,
                      verbose=2,
                      validation_split=VALIDATION_SPLIT)

Epoch 1/20
40/40 - 2s - loss: 0.6603 - accuracy: 0.6882 - val_loss: 0.6119 - val_accuracy: 0.7097
Epoch 2/20
40/40 - 0s - loss: 0.5424 - accuracy: 0.7476 - val_loss: 0.5103 - val_accuracy: 0.7388
Epoch 3/20
40/40 - 0s - loss: 0.4676 - accuracy: 0.7734 - val_loss: 0.4780 - val_accuracy: 0.7695
Epoch 4/20
40/40 - 1s - loss: 0.4499 - accuracy: 0.7728 - val_loss: 0.4729 - val_accuracy: 0.7718
Epoch 5/20
40/40 - 0s - loss: 0.4435 - accuracy: 0.7797 - val_loss: 0.4655 - val_accuracy: 0.7742
Epoch 6/20
40/40 - 0s - loss: 0.4383 - accuracy: 0.7886 - val_loss: 0.4591 - val_accuracy: 0.7671
Epoch 7/20
40/40 - 0s - loss: 0.4356 - accuracy: 0.7836 - val_loss: 0.4586 - val_accuracy: 0.7789
Epoch 8/20
40/40 - 0s - loss: 0.4288 - accuracy: 0.7909 - val_loss: 0.4506 - val_accuracy: 0.7742
Epoch 9/20
40/40 - 0s - loss: 0.4252 - accuracy: 0.7935 - val_loss: 0.4483 - val_accuracy: 0.7773
Epoch 10/20
40/40 - 0s - loss: 0.4237 - accuracy: 0.7952 - val_loss: 0.4483 - val_accuracy: 0.7797
Epoch 11/20
40/40 - 0s - loss: 0.4206 - accuracy: 0.7931 - val_loss: 0.4426 - val_accuracy: 0.7813
Epoch 12/20
40/40 - 0s - loss: 0.4169 - accuracy: 0.7982 - val_loss: 0.4449 - val_accuracy: 0.7860
Epoch 13/20
40/40 - 0s - loss: 0.4194 - accuracy: 0.7925 - val_loss: 0.4411 - val_accuracy: 0.7931
Epoch 14/20
40/40 - 0s - loss: 0.4134 - accuracy: 0.7976 - val_loss: 0.4390 - val_accuracy: 0.7899
Epoch 15/20
40/40 - 0s - loss: 0.4114 - accuracy: 0.8000 - val_loss: 0.4386 - val_accuracy: 0.7860
Epoch 16/20
40/40 - 0s - loss: 0.4122 - accuracy: 0.8007 - val_loss: 0.4390 - val_accuracy: 0.7836
Epoch 17/20
40/40 - 0s - loss: 0.4107 - accuracy: 0.8029 - val_loss: 0.4366 - val_accuracy: 0.7915
Epoch 18/20
40/40 - 0s - loss: 0.4126 - accuracy: 0.8025 - val_loss: 0.4332 - val_accuracy: 0.7891
Epoch 19/20
40/40 - 0s - loss: 0.4087 - accuracy: 0.8011 - val_loss: 0.4348 - val_accuracy: 0.7931
Epoch 20/20
40/40 - 0s - loss: 0.4055 - accuracy: 0.8047 - val_loss: 0.4336 - val_accuracy: 0.7962

plot1(history3)
../_images/dl-simple-case_73_0.png ../_images/dl-simple-case_73_1.png 
model3.evaluate(X_test, y_test, batch_size=128, verbose=2)

13/13 - 0s - loss: 0.4313 - accuracy: 0.7867

[0.4312911629676819, 0.7866582870483398]

Model 4: Improve the Models

  • In addition to regularization and dropouts, we can further improve the model by increasing the model complexity.

  • In particular, we can increase the depths and widths of the network layers.

  • Let’s try stacking two RNN layers.

Tip

When we stack two sequence layers (e.g., RNN), we need to make sure that the hidden states (outputs) of the first sequence layer at all timesteps are properly passed onto the next sequence layer, not just the hidden state (output) of the last timestep.

In keras, this usually means that we need to set the argument return_sequences=True in a sequence layer (e.g., SimpleRNN, LSTM, GRU etc).

## Define embedding dimension
MBEDDING_DIM = 128

## Define model
model4 = Sequential()
model4.add(
    layers.Embedding(input_dim=vocab_size,
                     output_dim=EMBEDDING_DIM,
                     input_length=max_len,
                     mask_zero=True))
model4.add(
    layers.SimpleRNN(16,
                     activation="relu",
                     name="RNN_layer_1",
                     dropout=0.2,
                     recurrent_dropout=0.2,
                     return_sequences=True) 
)  ## To ensure the hidden states of all timesteps are pased down to next layer
model4.add(
    layers.SimpleRNN(16,
                     activation="relu",
                     name="RNN_layer_2",
                     dropout=0.2,
                     recurrent_dropout=0.2))
model4.add(layers.Dense(1, activation="sigmoid", name="output"))

## Compile model
model4.compile(loss=keras.losses.BinaryCrossentropy(),
               optimizer=keras.optimizers.Adam(lr=0.001),
               metrics=["accuracy"])

plot_model(model4, show_shapes=True)
../_images/dl-simple-case_79_0.png 
history4 = model4.fit(X_train,
                      y_train,
                      batch_size=BATCH_SIZE,
                      epochs=EPOCHS,
                      verbose=2,
                      validation_split=VALIDATION_SPLIT)

Epoch 1/20
40/40 - 4s - loss: 0.6493 - accuracy: 0.6239 - val_loss: 0.6130 - val_accuracy: 0.6310
Epoch 2/20
40/40 - 0s - loss: 0.5805 - accuracy: 0.6951 - val_loss: 0.5246 - val_accuracy: 0.7223
Epoch 3/20
40/40 - 1s - loss: 0.5190 - accuracy: 0.7425 - val_loss: 0.4749 - val_accuracy: 0.7710
Epoch 4/20
40/40 - 0s - loss: 0.4791 - accuracy: 0.7685 - val_loss: 0.4595 - val_accuracy: 0.7797
Epoch 5/20
40/40 - 1s - loss: 0.4658 - accuracy: 0.7754 - val_loss: 0.4553 - val_accuracy: 0.7852
Epoch 6/20
40/40 - 0s - loss: 0.4602 - accuracy: 0.7742 - val_loss: 0.4517 - val_accuracy: 0.7844
Epoch 7/20
40/40 - 0s - loss: 0.4493 - accuracy: 0.7744 - val_loss: 0.4484 - val_accuracy: 0.7884
Epoch 8/20
40/40 - 0s - loss: 0.4447 - accuracy: 0.7799 - val_loss: 0.4501 - val_accuracy: 0.7844
Epoch 9/20
40/40 - 0s - loss: 0.4395 - accuracy: 0.7891 - val_loss: 0.4470 - val_accuracy: 0.7876
Epoch 10/20
40/40 - 0s - loss: 0.4370 - accuracy: 0.7931 - val_loss: 0.4410 - val_accuracy: 0.7954
Epoch 11/20
40/40 - 1s - loss: 0.4327 - accuracy: 0.7939 - val_loss: 0.4389 - val_accuracy: 0.7931
Epoch 12/20
40/40 - 1s - loss: 0.4298 - accuracy: 0.7848 - val_loss: 0.4375 - val_accuracy: 0.7939
Epoch 13/20
40/40 - 0s - loss: 0.4352 - accuracy: 0.7899 - val_loss: 0.4387 - val_accuracy: 0.7931
Epoch 14/20
40/40 - 1s - loss: 0.4265 - accuracy: 0.7990 - val_loss: 0.4432 - val_accuracy: 0.7907
Epoch 15/20
40/40 - 1s - loss: 0.4277 - accuracy: 0.7954 - val_loss: 0.4358 - val_accuracy: 0.7978
Epoch 16/20
40/40 - 1s - loss: 0.4263 - accuracy: 0.7964 - val_loss: 0.4374 - val_accuracy: 0.8009
Epoch 17/20
40/40 - 1s - loss: 0.4238 - accuracy: 0.7966 - val_loss: 0.4353 - val_accuracy: 0.7954
Epoch 18/20
40/40 - 1s - loss: 0.4189 - accuracy: 0.8031 - val_loss: 0.4345 - val_accuracy: 0.7931
Epoch 19/20
40/40 - 1s - loss: 0.4246 - accuracy: 0.7952 - val_loss: 0.4357 - val_accuracy: 0.7884
Epoch 20/20
40/40 - 1s - loss: 0.4203 - accuracy: 0.8065 - val_loss: 0.4376 - val_accuracy: 0.7931

plot1(history4)
../_images/dl-simple-case_81_0.png ../_images/dl-simple-case_81_1.png 
model4.evaluate(X_test, y_test, batch_size=128, verbose=2)

13/13 - 0s - loss: 0.4272 - accuracy: 0.7904

[0.4271852970123291, 0.7904342412948608]

Model 5: Bidirectional

  • We can also increase the model complexity in at least two possible ways:

    • Use more advanced RNNs, such as LSTM or GRU

    • Process the sequence in two directions

    • Increase the hidden nodes of the RNN/LSTM

  • Now let’s try the more sophisticated RNN, LSTM, and with bidirectional sequence processing and add more nodes to the LSTM layer.

## Define embedding dimension
EMBEDDING_DIM = 128

## Define model
model5 = Sequential()
model5.add(
    layers.Embedding(input_dim=vocab_size,
                      output_dim=EMBEDDING_DIM,
                      input_length=max_len,
                      mask_zero=True))
model5.add(
    layers.Bidirectional(  ## Bidirectional sequence processing
        layers.LSTM(32,
                    activation="relu",
                    name="lstm_layer_1",
                    dropout=0.2,
                    recurrent_dropout=0.5,
                    return_sequences=True)))
model5.add(
    layers.Bidirectional(  ## Bidirectional sequence processing
        layers.LSTM(32,
                    activation="relu",
                    name="lstm_layer_2",
                    dropout=0.2,
                    recurrent_dropout=0.5)))
model5.add(layers.Dense(1, activation="sigmoid", name="output"))

model5.compile(loss=keras.losses.BinaryCrossentropy(),
               optimizer=keras.optimizers.Adam(lr=0.001),
               metrics=["accuracy"])

plot_model(model5, show_shapes=True)
../_images/dl-simple-case_86_0.png 
history5 = model5.fit(X_train,
                      y_train,
                      batch_size=BATCH_SIZE,
                      epochs=EPOCHS,
                      verbose=2,
                      validation_split=VALIDATION_SPLIT)

Epoch 1/20
40/40 - 16s - loss: 0.6629 - accuracy: 0.6220 - val_loss: 0.6377 - val_accuracy: 0.6121
Epoch 2/20
40/40 - 2s - loss: 0.5933 - accuracy: 0.6690 - val_loss: 0.5413 - val_accuracy: 0.7396
Epoch 3/20
40/40 - 2s - loss: 0.5076 - accuracy: 0.7455 - val_loss: 0.5043 - val_accuracy: 0.7655
Epoch 4/20
40/40 - 3s - loss: 0.4815 - accuracy: 0.7695 - val_loss: 0.4847 - val_accuracy: 0.7773
Epoch 5/20
40/40 - 2s - loss: 0.4662 - accuracy: 0.7730 - val_loss: 0.4731 - val_accuracy: 0.7821
Epoch 6/20
40/40 - 2s - loss: 0.4515 - accuracy: 0.7854 - val_loss: 0.4608 - val_accuracy: 0.7836
Epoch 7/20
40/40 - 2s - loss: 0.4423 - accuracy: 0.7817 - val_loss: 0.4554 - val_accuracy: 0.7836
Epoch 8/20
40/40 - 2s - loss: 0.4323 - accuracy: 0.7917 - val_loss: 0.4664 - val_accuracy: 0.7789
Epoch 9/20
40/40 - 2s - loss: 0.4277 - accuracy: 0.7929 - val_loss: 0.4445 - val_accuracy: 0.7844
Epoch 10/20
40/40 - 2s - loss: 0.4168 - accuracy: 0.7968 - val_loss: 0.4372 - val_accuracy: 0.7868
Epoch 11/20
40/40 - 2s - loss: 0.4110 - accuracy: 0.8006 - val_loss: 0.4341 - val_accuracy: 0.7962
Epoch 12/20
40/40 - 2s - loss: 0.4103 - accuracy: 0.8063 - val_loss: 0.4453 - val_accuracy: 0.7931
Epoch 13/20
40/40 - 2s - loss: 0.4047 - accuracy: 0.8078 - val_loss: 0.4356 - val_accuracy: 0.7994
Epoch 14/20
40/40 - 3s - loss: 0.3998 - accuracy: 0.8051 - val_loss: 0.4310 - val_accuracy: 0.7970
Epoch 15/20
40/40 - 3s - loss: 0.3990 - accuracy: 0.8118 - val_loss: 0.4310 - val_accuracy: 0.7970
Epoch 16/20
40/40 - 2s - loss: 0.3983 - accuracy: 0.8066 - val_loss: 0.4319 - val_accuracy: 0.8017
Epoch 17/20
40/40 - 3s - loss: 0.3972 - accuracy: 0.8125 - val_loss: 0.4235 - val_accuracy: 0.7994
Epoch 18/20
40/40 - 3s - loss: 0.3960 - accuracy: 0.8102 - val_loss: 0.4208 - val_accuracy: 0.8033
Epoch 19/20
40/40 - 3s - loss: 0.3949 - accuracy: 0.8112 - val_loss: 0.4199 - val_accuracy: 0.8009
Epoch 20/20
40/40 - 3s - loss: 0.3867 - accuracy: 0.8204 - val_loss: 0.4224 - val_accuracy: 0.8033

plot1(history5)
../_images/dl-simple-case_88_0.png ../_images/dl-simple-case_88_1.png 
model5.evaluate(X_test, y_test, batch_size=128, verbose=2)

13/13 - 0s - loss: 0.4200 - accuracy: 0.8024

[0.4199684262275696, 0.8023914694786072]

2.8. Check Embeddings

  • Compared to one-hot encodings of characters, embeddings may include more information relating to the characteristics (semantics?) of the characters.

  • We can extract the embedding layer and apply dimensional reduction techniques (i.e., TSNE) to see how embeddings capture the relationships in-between characters.

## A name in sequence from test set
print(X_test_texts[10])
print(X_test[10])

Erma
[ 0  0  0  0  0  0  0  0  0  0  0  2  5 12  1]

## Extract Corpus Dictionary (mapping of chars and integer indices)
ind2char = tokenizer.index_word
[ind2char.get(i) for i in X_test[10] if ind2char.get(i) != None]

['e', 'r', 'm', 'a']

## Extract the embedding layer (its weights matrix)
char_vectors = model5.layers[0].get_weights()[0]
print(char_vectors.shape) ## embedding shape (vocab_size, embed_dim)
print(char_vectors[1,:]) ## first char embeddings

(30, 128)
[-0.08638821  0.10196032 -0.08930968 -0.08678097 -0.09036665 -0.06538443
  0.06650709 -0.10806406  0.00168613 -0.07606699 -0.09379511 -0.05240235
 -0.06894766  0.01251148  0.23080757 -0.12168883 -0.11802603 -0.12854312
  0.12610877 -0.01837258 -0.08400774  0.06846871 -0.07555567 -0.07070152
 -0.08286686  0.10378125  0.09733976 -0.09550156  0.02652814  0.06678072
  0.09279401 -0.00279973 -0.14800397 -0.09717548 -0.02988891  0.20770137
 -0.10793544 -0.13713866 -0.10818603  0.05101357 -0.08654831 -0.10884775
  0.10200328 -0.05336187 -0.10099219  0.0701826   0.06384768  0.09715796
 -0.12641913 -0.06987549  0.09652784  0.03662771 -0.04180462 -0.04470819
 -0.1648816  -0.16170195  0.13866846  0.03711467 -0.18266453  0.12483152
  0.09363896 -0.07205503 -0.10150398  0.13476276 -0.09745233  0.12633258
  0.0910398   0.08862504 -0.19972621  0.07722508  0.1255034   0.14225055
  0.12625924  0.12086959 -0.09569547 -0.11019094  0.12148605 -0.09907743
  0.0770528  -0.10457225  0.19166091  0.10602122  0.04267744 -0.07552098
  0.2044732   0.12589149  0.15693201 -0.1005476  -0.14037779 -0.1188178
 -0.04631033 -0.12802887 -0.10465906 -0.1166728   0.05874987 -0.11016585
  0.10345474  0.01566136  0.11203294 -0.00494666  0.08957522 -0.05410516
 -0.08413711 -0.16119377 -0.08949109 -0.11353346  0.1060674   0.14916374
 -0.09312931 -0.01845127 -0.10890387 -0.0339985  -0.08260787 -0.08899793
 -0.10548896 -0.07199826  0.09011649  0.05084977  0.03348807  0.05117476
 -0.14368454 -0.02789242  0.15400283 -0.04970378 -0.13955727  0.04079244
  0.02812686  0.05327073]

labels = [char for (ind, char) in tokenizer.index_word.items()]
labels.insert(0, None)
labels

[None,
 'a',
 'e',
 'i',
 'n',
 'r',
 'l',
 'o',
 't',
 's',
 'd',
 'y',
 'm',
 'h',
 'c',
 'b',
 'u',
 'g',
 'k',
 'j',
 'v',
 'f',
 'p',
 'w',
 'z',
 'x',
 'q',
 '-',
 ' ',
 "'"]

## Visulizing char embeddings via dimensional reduction techniques

tsne = TSNE(n_components=2, random_state=0, n_iter=5000, perplexity=3)
np.set_printoptions(suppress=True)

T = tsne.fit_transform(char_vectors)
labels = labels

plt.figure(figsize=(10, 7), dpi=150)
plt.scatter(T[:, 0], T[:, 1], c='orange', edgecolors='r')
for label, x, y in zip(labels, T[:, 0], T[:, 1]):
    plt.annotate(label,
                 xy=(x + 1, y + 1),
                 xytext=(0, 0),
                 textcoords='offset points')
../_images/dl-simple-case_96_0.png

2.9. Issues of Word/Character Representations

  • One-hot encoding does not indicate semantic relationships between characters.

  • For deep learning NLP, it is preferred to convert one-hot encodings of words/characters into embeddings, which are argued to include more semantic information of the tokens.

  • Now the question is how to train and create better word embeddings.

  • There are at least two alternatives:

    • We can train the embeddings along with our current NLP task.

    • We can use pre-trained embeddings from other unsupervised learning (transfer learning).

  • We will come back to this issue later.

2.10. Hyperparameter Tuning

Note

Please install keras tuner module in your current conda:

pip install -U keras-tuner

or

conda install -c conda-forge keras-tuner

  • Like feature-based ML methods, neural networks also come with many hyperparameters, which require default values.

  • Typical hyperparameters include:

    • Number of nodes for the layer

    • Learning Rates

  • We can utilize the module, keras-tuner, to fine-tune these hyperparameters (i.e., to find the values that optimize the model performance).

  • Steps for Keras Tuner

    • First, wrap the model definition in a function, which takes a single hp argument.

    • Inside this function, replace any value we want to tune with a call to hyperparameter sampling methods, e.g. hp.Int() or hp.Choice(). The function should return a compiled model.

    • Next, instantiate a tuner object specifying our optimization objective and other search parameters.

    • Finally, start the search with the search() method, which takes the same arguments as Model.fit() in keras.

    • When the search is over, we can retrieve the best model and a summary of the results from the tunner.

## confirm if the right kernel is being used
# import sys
# sys.executable

## Wrap model definition in a function
## and specify the parameters needed for tuning
# def build_model(hp):
#     model1 = keras.Sequential()
#     model1.add(keras.Input(shape=(max_len,)))
#     model1.add(layers.Dense(hp.Int('units', min_value=32, max_value=128, step=32), activation="relu", name="dense_layer_1"))
#     model1.add(layers.Dense(hp.Int('units', min_value=32, max_value=128, step=32), activation="relu", name="dense_layer_2"))
#     model1.add(layers.Dense(2, activation="softmax", name="output"))
#     model1.compile(
#         optimizer=keras.optimizers.Adam(
#             hp.Choice('learning_rate',
#                       values=[1e-2, 1e-3, 1e-4])),
#         loss='sparse_categorical_crossentropy',
#         metrics=['accuracy'])
#     return model1

## wrap model definition and compiling

def build_model(hp):
    m = Sequential()
    m.add(
        layers.Embedding(
            input_dim=vocab_size,
            output_dim=hp.Int(
                'output_dim',  ## tuning 2
                min_value=32,
                max_value=128,
                step=32),
            input_length=max_len,
            mask_zero=True))
    m.add(
        layers.Bidirectional(
            layers.LSTM(
                hp.Int('units', min_value=16, max_value=64,
                       step=16),  ## tuning 1
                activation="relu",
                dropout=0.2,
                recurrent_dropout=0.2)))
    m.add(layers.Dense(1, activation="sigmoid", name="output"))

    m.compile(loss=keras.losses.BinaryCrossentropy(),
              optimizer=keras.optimizers.Adam(lr=0.001),
              metrics=["accuracy"])
    return m

## This is to clean up the temp dir from the tuner
## Every time we re-start the tunner, it's better to keep the temp dir clean

if os.path.isdir('my_dir'):
    shutil.rmtree('my_dir')

  • The max_trials variable represents the maximum number of trials that a hyperparameter combination would run.

  • The execution_per_trial variable is the number of models that should be built and fit for each trial for robustness purposes.

## Instantiate the tunner

tuner = kerastuner.tuners.RandomSearch(build_model,
                                       objective='val_accuracy',
                                       max_trials=10,
                                       executions_per_trial=2,
                                       directory='my_dir')

## Check the tuner's search space
tuner.search_space_summary()

Search space summary

|-Default search space size: 2

output_dim (Int)

|-default: None
|-max_value: 128
|-min_value: 32
|-sampling: None
|-step: 32

units (Int)

|-default: None
|-max_value: 64
|-min_value: 16
|-sampling: None
|-step: 16
%%time
## Start tuning with the tuner
tuner.search(X_train, y_train, validation_split=0.2, batch_size=128)

40/40 [==============================] - ETA: 3:51 - loss: 0.6929 - accuracy: 0.55 - ETA: 1s - loss: 0.6919 - accuracy: 0.5855 - ETA: 0s - loss: 0.6906 - accuracy: 0.60 - ETA: 0s - loss: 0.6896 - accuracy: 0.61 - ETA: 0s - loss: 0.6887 - accuracy: 0.61 - ETA: 0s - loss: 0.6879 - accuracy: 0.62 - ETA: 0s - loss: 0.6869 - accuracy: 0.62 - ETA: 0s - loss: 0.6859 - accuracy: 0.62 - ETA: 0s - loss: 0.6852 - accuracy: 0.62 - ETA: 0s - loss: 0.6845 - accuracy: 0.62 - ETA: 0s - loss: 0.6835 - accuracy: 0.62 - ETA: 0s - loss: 0.6828 - accuracy: 0.62 - ETA: 0s - loss: 0.6821 - accuracy: 0.62 - ETA: 0s - loss: 0.6814 - accuracy: 0.62 - ETA: 0s - loss: 0.6807 - accuracy: 0.62 - ETA: 0s - loss: 0.6800 - accuracy: 0.62 - ETA: 0s - loss: 0.6792 - accuracy: 0.62 - ETA: 0s - loss: 0.6784 - accuracy: 0.62 - 8s 56ms/step - loss: 0.6775 - accuracy: 0.6249 - val_loss: 0.6414 - val_accuracy: 0.6121
40/40 [==============================] - ETA: 3:47 - loss: 0.6913 - accuracy: 0.71 - ETA: 1s - loss: 0.6904 - accuracy: 0.6966 - ETA: 1s - loss: 0.6899 - accuracy: 0.68 - ETA: 1s - loss: 0.6893 - accuracy: 0.67 - ETA: 1s - loss: 0.6887 - accuracy: 0.66 - ETA: 1s - loss: 0.6881 - accuracy: 0.65 - ETA: 1s - loss: 0.6874 - accuracy: 0.65 - ETA: 0s - loss: 0.6867 - accuracy: 0.65 - ETA: 0s - loss: 0.6859 - accuracy: 0.65 - ETA: 0s - loss: 0.6855 - accuracy: 0.64 - ETA: 0s - loss: 0.6848 - accuracy: 0.64 - ETA: 0s - loss: 0.6840 - accuracy: 0.64 - ETA: 0s - loss: 0.6834 - accuracy: 0.64 - ETA: 0s - loss: 0.6827 - accuracy: 0.64 - ETA: 0s - loss: 0.6819 - accuracy: 0.64 - ETA: 0s - loss: 0.6812 - accuracy: 0.64 - ETA: 0s - loss: 0.6803 - accuracy: 0.64 - ETA: 0s - loss: 0.6795 - accuracy: 0.63 - ETA: 0s - loss: 0.6787 - accuracy: 0.63 - ETA: 0s - loss: 0.6783 - accuracy: 0.63 - ETA: 0s - loss: 0.6778 - accuracy: 0.63 - ETA: 0s - loss: 0.6769 - accuracy: 0.63 - 8s 62ms/step - loss: 0.6765 - accuracy: 0.6370 - val_loss: 0.6397 - val_accuracy: 0.6121

Trial complete

Trial summary

|-Trial ID: 193383d74737cdad5c526eba4cf2dc03
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 64
|-units: 32
40/40 [==============================] - ETA: 3:43 - loss: 0.6926 - accuracy: 0.51 - ETA: 1s - loss: 0.6919 - accuracy: 0.5430 - ETA: 1s - loss: 0.6908 - accuracy: 0.56 - ETA: 0s - loss: 0.6899 - accuracy: 0.57 - ETA: 0s - loss: 0.6891 - accuracy: 0.58 - ETA: 0s - loss: 0.6883 - accuracy: 0.58 - ETA: 0s - loss: 0.6874 - accuracy: 0.59 - ETA: 0s - loss: 0.6865 - accuracy: 0.59 - ETA: 0s - loss: 0.6861 - accuracy: 0.59 - ETA: 0s - loss: 0.6850 - accuracy: 0.59 - ETA: 0s - loss: 0.6840 - accuracy: 0.60 - ETA: 0s - loss: 0.6830 - accuracy: 0.60 - ETA: 0s - loss: 0.6819 - accuracy: 0.60 - ETA: 0s - loss: 0.6809 - accuracy: 0.60 - ETA: 0s - loss: 0.6798 - accuracy: 0.60 - ETA: 0s - loss: 0.6787 - accuracy: 0.61 - ETA: 0s - loss: 0.6778 - accuracy: 0.61 - ETA: 0s - loss: 0.6769 - accuracy: 0.61 - ETA: 0s - loss: 0.6760 - accuracy: 0.61 - ETA: 0s - loss: 0.6751 - accuracy: 0.61 - ETA: 0s - loss: 0.6747 - accuracy: 0.61 - 8s 54ms/step - loss: 0.6740 - accuracy: 0.6148 - val_loss: 0.6362 - val_accuracy: 0.6121
40/40 [==============================] - ETA: 3:46 - loss: 0.6916 - accuracy: 0.65 - ETA: 1s - loss: 0.6912 - accuracy: 0.6341 - ETA: 1s - loss: 0.6904 - accuracy: 0.63 - ETA: 1s - loss: 0.6897 - accuracy: 0.62 - ETA: 1s - loss: 0.6888 - accuracy: 0.62 - ETA: 0s - loss: 0.6878 - accuracy: 0.62 - ETA: 0s - loss: 0.6869 - accuracy: 0.62 - ETA: 0s - loss: 0.6864 - accuracy: 0.62 - ETA: 0s - loss: 0.6859 - accuracy: 0.62 - ETA: 0s - loss: 0.6854 - accuracy: 0.62 - ETA: 0s - loss: 0.6845 - accuracy: 0.63 - ETA: 0s - loss: 0.6835 - accuracy: 0.63 - ETA: 0s - loss: 0.6826 - accuracy: 0.63 - ETA: 0s - loss: 0.6816 - accuracy: 0.63 - ETA: 0s - loss: 0.6807 - accuracy: 0.63 - ETA: 0s - loss: 0.6798 - accuracy: 0.63 - ETA: 0s - loss: 0.6789 - accuracy: 0.63 - ETA: 0s - loss: 0.6781 - accuracy: 0.63 - ETA: 0s - loss: 0.6771 - accuracy: 0.63 - ETA: 0s - loss: 0.6762 - accuracy: 0.63 - ETA: 0s - loss: 0.6753 - accuracy: 0.63 - 8s 56ms/step - loss: 0.6741 - accuracy: 0.6302 - val_loss: 0.6350 - val_accuracy: 0.6121

Trial complete

Trial summary

|-Trial ID: f169fc59e0f12c5e0dad3a1304f88e86
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 64
|-units: 48
40/40 [==============================] - ETA: 4:05 - loss: 0.6915 - accuracy: 0.64 - ETA: 1s - loss: 0.6912 - accuracy: 0.6263 - ETA: 0s - loss: 0.6907 - accuracy: 0.61 - ETA: 0s - loss: 0.6900 - accuracy: 0.61 - ETA: 0s - loss: 0.6895 - accuracy: 0.61 - ETA: 0s - loss: 0.6890 - accuracy: 0.61 - ETA: 0s - loss: 0.6885 - accuracy: 0.61 - ETA: 0s - loss: 0.6880 - accuracy: 0.61 - ETA: 0s - loss: 0.6876 - accuracy: 0.61 - ETA: 0s - loss: 0.6871 - accuracy: 0.61 - ETA: 0s - loss: 0.6864 - accuracy: 0.61 - ETA: 0s - loss: 0.6859 - accuracy: 0.61 - ETA: 0s - loss: 0.6853 - accuracy: 0.61 - ETA: 0s - loss: 0.6847 - accuracy: 0.61 - ETA: 0s - loss: 0.6841 - accuracy: 0.61 - ETA: 0s - loss: 0.6834 - accuracy: 0.61 - ETA: 0s - loss: 0.6827 - accuracy: 0.62 - ETA: 0s - loss: 0.6820 - accuracy: 0.62 - 8s 49ms/step - loss: 0.6810 - accuracy: 0.6213 - val_loss: 0.6503 - val_accuracy: 0.6121
40/40 [==============================] - ETA: 4:06 - loss: 0.6930 - accuracy: 0.53 - ETA: 0s - loss: 0.6925 - accuracy: 0.5799 - ETA: 0s - loss: 0.6919 - accuracy: 0.59 - ETA: 0s - loss: 0.6913 - accuracy: 0.60 - ETA: 0s - loss: 0.6906 - accuracy: 0.60 - ETA: 0s - loss: 0.6899 - accuracy: 0.61 - ETA: 0s - loss: 0.6892 - accuracy: 0.61 - ETA: 0s - loss: 0.6887 - accuracy: 0.61 - ETA: 0s - loss: 0.6878 - accuracy: 0.61 - ETA: 0s - loss: 0.6870 - accuracy: 0.61 - ETA: 0s - loss: 0.6864 - accuracy: 0.61 - ETA: 0s - loss: 0.6861 - accuracy: 0.61 - ETA: 0s - loss: 0.6855 - accuracy: 0.61 - ETA: 0s - loss: 0.6849 - accuracy: 0.61 - ETA: 0s - loss: 0.6840 - accuracy: 0.61 - ETA: 0s - loss: 0.6834 - accuracy: 0.62 - ETA: 0s - loss: 0.6828 - accuracy: 0.62 - 8s 47ms/step - loss: 0.6822 - accuracy: 0.6205 - val_loss: 0.6510 - val_accuracy: 0.6121

Trial complete

Trial summary

|-Trial ID: 0c3a82a52f5438822e0f94f232c7203e
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 32
|-units: 32
40/40 [==============================] - ETA: 3:53 - loss: 0.6939 - accuracy: 0.43 - ETA: 2s - loss: 0.6935 - accuracy: 0.4746 - ETA: 2s - loss: 0.6932 - accuracy: 0.49 - ETA: 1s - loss: 0.6926 - accuracy: 0.52 - ETA: 1s - loss: 0.6919 - accuracy: 0.54 - ETA: 1s - loss: 0.6913 - accuracy: 0.55 - ETA: 1s - loss: 0.6907 - accuracy: 0.56 - ETA: 1s - loss: 0.6900 - accuracy: 0.57 - ETA: 1s - loss: 0.6897 - accuracy: 0.57 - ETA: 0s - loss: 0.6889 - accuracy: 0.58 - ETA: 0s - loss: 0.6881 - accuracy: 0.58 - ETA: 0s - loss: 0.6877 - accuracy: 0.58 - ETA: 0s - loss: 0.6873 - accuracy: 0.59 - ETA: 0s - loss: 0.6865 - accuracy: 0.59 - ETA: 0s - loss: 0.6858 - accuracy: 0.59 - ETA: 0s - loss: 0.6851 - accuracy: 0.59 - ETA: 0s - loss: 0.6845 - accuracy: 0.60 - ETA: 0s - loss: 0.6838 - accuracy: 0.60 - ETA: 0s - loss: 0.6831 - accuracy: 0.60 - ETA: 0s - loss: 0.6824 - accuracy: 0.60 - ETA: 0s - loss: 0.6817 - accuracy: 0.60 - ETA: 0s - loss: 0.6810 - accuracy: 0.60 - ETA: 0s - loss: 0.6803 - accuracy: 0.60 - 8s 58ms/step - loss: 0.6800 - accuracy: 0.6081 - val_loss: 0.6492 - val_accuracy: 0.6121
40/40 [==============================] - ETA: 3:42 - loss: 0.6926 - accuracy: 0.57 - ETA: 1s - loss: 0.6919 - accuracy: 0.6102 - ETA: 1s - loss: 0.6912 - accuracy: 0.61 - ETA: 0s - loss: 0.6907 - accuracy: 0.62 - ETA: 0s - loss: 0.6901 - accuracy: 0.62 - ETA: 0s - loss: 0.6896 - accuracy: 0.62 - ETA: 0s - loss: 0.6890 - accuracy: 0.62 - ETA: 0s - loss: 0.6884 - accuracy: 0.62 - ETA: 0s - loss: 0.6878 - accuracy: 0.62 - ETA: 0s - loss: 0.6873 - accuracy: 0.62 - ETA: 0s - loss: 0.6866 - accuracy: 0.62 - ETA: 0s - loss: 0.6860 - accuracy: 0.62 - ETA: 0s - loss: 0.6854 - accuracy: 0.62 - ETA: 0s - loss: 0.6847 - accuracy: 0.62 - ETA: 0s - loss: 0.6840 - accuracy: 0.62 - ETA: 0s - loss: 0.6833 - accuracy: 0.62 - ETA: 0s - loss: 0.6821 - accuracy: 0.62 - ETA: 0s - loss: 0.6813 - accuracy: 0.62 - ETA: 0s - loss: 0.6805 - accuracy: 0.62 - ETA: 0s - loss: 0.6796 - accuracy: 0.62 - 8s 49ms/step - loss: 0.6793 - accuracy: 0.6253 - val_loss: 0.6512 - val_accuracy: 0.6121

Trial complete

Trial summary

|-Trial ID: 3cf4f74b266185196862938ef406c47a
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 32
|-units: 48
40/40 [==============================] - ETA: 3:49 - loss: 0.6939 - accuracy: 0.46 - ETA: 1s - loss: 0.6931 - accuracy: 0.5065 - ETA: 0s - loss: 0.6920 - accuracy: 0.54 - ETA: 0s - loss: 0.6909 - accuracy: 0.56 - ETA: 0s - loss: 0.6901 - accuracy: 0.57 - ETA: 0s - loss: 0.6892 - accuracy: 0.58 - ETA: 0s - loss: 0.6884 - accuracy: 0.59 - ETA: 0s - loss: 0.6872 - accuracy: 0.59 - ETA: 0s - loss: 0.6865 - accuracy: 0.59 - ETA: 0s - loss: 0.6858 - accuracy: 0.60 - ETA: 0s - loss: 0.6850 - accuracy: 0.60 - ETA: 0s - loss: 0.6842 - accuracy: 0.60 - ETA: 0s - loss: 0.6834 - accuracy: 0.60 - ETA: 0s - loss: 0.6826 - accuracy: 0.60 - ETA: 0s - loss: 0.6817 - accuracy: 0.60 - ETA: 0s - loss: 0.6808 - accuracy: 0.60 - ETA: 0s - loss: 0.6798 - accuracy: 0.60 - ETA: 0s - loss: 0.6788 - accuracy: 0.61 - ETA: 0s - loss: 0.6778 - accuracy: 0.61 - ETA: 0s - loss: 0.6769 - accuracy: 0.61 - 8s 49ms/step - loss: 0.6764 - accuracy: 0.6122 - val_loss: 0.6398 - val_accuracy: 0.6121
40/40 [==============================] - ETA: 3:33 - loss: 0.6926 - accuracy: 0.60 - ETA: 2s - loss: 0.6920 - accuracy: 0.6113 - ETA: 1s - loss: 0.6909 - accuracy: 0.62 - ETA: 1s - loss: 0.6897 - accuracy: 0.63 - ETA: 1s - loss: 0.6887 - accuracy: 0.63 - ETA: 1s - loss: 0.6878 - accuracy: 0.63 - ETA: 0s - loss: 0.6868 - accuracy: 0.63 - ETA: 0s - loss: 0.6858 - accuracy: 0.63 - ETA: 0s - loss: 0.6847 - accuracy: 0.63 - ETA: 0s - loss: 0.6833 - accuracy: 0.63 - ETA: 0s - loss: 0.6819 - accuracy: 0.63 - ETA: 0s - loss: 0.6809 - accuracy: 0.63 - ETA: 0s - loss: 0.6798 - accuracy: 0.63 - ETA: 0s - loss: 0.6787 - accuracy: 0.63 - ETA: 0s - loss: 0.6777 - accuracy: 0.63 - ETA: 0s - loss: 0.6766 - accuracy: 0.63 - ETA: 0s - loss: 0.6756 - accuracy: 0.63 - ETA: 0s - loss: 0.6742 - accuracy: 0.63 - ETA: 0s - loss: 0.6732 - accuracy: 0.63 - 7s 47ms/step - loss: 0.6724 - accuracy: 0.6351 - val_loss: 0.6273 - val_accuracy: 0.6137

Trial complete

Trial summary

|-Trial ID: 78c668676dde2b55ddda6af9c183580c
|-Score: 0.6129032373428345
|-Best step: 0

Hyperparameters:

|-output_dim: 96
|-units: 32
40/40 [==============================] - ETA: 3:35 - loss: 0.6960 - accuracy: 0.34 - ETA: 1s - loss: 0.6943 - accuracy: 0.4314 - ETA: 1s - loss: 0.6927 - accuracy: 0.48 - ETA: 1s - loss: 0.6912 - accuracy: 0.52 - ETA: 1s - loss: 0.6900 - accuracy: 0.54 - ETA: 1s - loss: 0.6889 - accuracy: 0.55 - ETA: 0s - loss: 0.6878 - accuracy: 0.56 - ETA: 0s - loss: 0.6868 - accuracy: 0.57 - ETA: 0s - loss: 0.6860 - accuracy: 0.57 - ETA: 0s - loss: 0.6851 - accuracy: 0.57 - ETA: 0s - loss: 0.6842 - accuracy: 0.58 - ETA: 0s - loss: 0.6832 - accuracy: 0.58 - ETA: 0s - loss: 0.6822 - accuracy: 0.58 - ETA: 0s - loss: 0.6817 - accuracy: 0.58 - ETA: 0s - loss: 0.6805 - accuracy: 0.59 - ETA: 0s - loss: 0.6794 - accuracy: 0.59 - ETA: 0s - loss: 0.6783 - accuracy: 0.59 - ETA: 0s - loss: 0.6771 - accuracy: 0.59 - ETA: 0s - loss: 0.6761 - accuracy: 0.59 - ETA: 0s - loss: 0.6751 - accuracy: 0.59 - ETA: 0s - loss: 0.6741 - accuracy: 0.59 - 8s 58ms/step - loss: 0.6737 - accuracy: 0.6004 - val_loss: 0.6231 - val_accuracy: 0.6176
40/40 [==============================] - ETA: 3:47 - loss: 0.6923 - accuracy: 0.57 - ETA: 1s - loss: 0.6915 - accuracy: 0.5903 - ETA: 1s - loss: 0.6903 - accuracy: 0.59 - ETA: 1s - loss: 0.6893 - accuracy: 0.60 - ETA: 1s - loss: 0.6880 - accuracy: 0.60 - ETA: 1s - loss: 0.6873 - accuracy: 0.61 - ETA: 1s - loss: 0.6866 - accuracy: 0.61 - ETA: 1s - loss: 0.6860 - accuracy: 0.61 - ETA: 1s - loss: 0.6846 - accuracy: 0.61 - ETA: 1s - loss: 0.6833 - accuracy: 0.61 - ETA: 0s - loss: 0.6821 - accuracy: 0.62 - ETA: 0s - loss: 0.6809 - accuracy: 0.62 - ETA: 0s - loss: 0.6796 - accuracy: 0.62 - ETA: 0s - loss: 0.6783 - accuracy: 0.62 - ETA: 0s - loss: 0.6772 - accuracy: 0.62 - ETA: 0s - loss: 0.6760 - accuracy: 0.62 - ETA: 0s - loss: 0.6749 - accuracy: 0.62 - ETA: 0s - loss: 0.6738 - accuracy: 0.62 - ETA: 0s - loss: 0.6727 - accuracy: 0.62 - ETA: 0s - loss: 0.6716 - accuracy: 0.62 - ETA: 0s - loss: 0.6705 - accuracy: 0.62 - ETA: 0s - loss: 0.6695 - accuracy: 0.62 - 8s 58ms/step - loss: 0.6690 - accuracy: 0.6265 - val_loss: 0.6250 - val_accuracy: 0.6161

Trial complete

Trial summary

|-Trial ID: ae64d7ed05ffee5ffebb5db7ff7da094
|-Score: 0.6168371438980103
|-Best step: 0

Hyperparameters:

|-output_dim: 128
|-units: 48
40/40 [==============================] - ETA: 3:37 - loss: 0.6943 - accuracy: 0.39 - ETA: 1s - loss: 0.6934 - accuracy: 0.4627 - ETA: 1s - loss: 0.6927 - accuracy: 0.49 - ETA: 0s - loss: 0.6917 - accuracy: 0.52 - ETA: 0s - loss: 0.6907 - accuracy: 0.53 - ETA: 0s - loss: 0.6897 - accuracy: 0.55 - ETA: 1s - loss: 0.6891 - accuracy: 0.55 - ETA: 1s - loss: 0.6887 - accuracy: 0.55 - ETA: 0s - loss: 0.6878 - accuracy: 0.56 - ETA: 0s - loss: 0.6868 - accuracy: 0.57 - ETA: 0s - loss: 0.6858 - accuracy: 0.57 - ETA: 0s - loss: 0.6848 - accuracy: 0.58 - ETA: 0s - loss: 0.6838 - accuracy: 0.58 - ETA: 0s - loss: 0.6827 - accuracy: 0.58 - ETA: 0s - loss: 0.6817 - accuracy: 0.59 - ETA: 0s - loss: 0.6807 - accuracy: 0.59 - ETA: 0s - loss: 0.6797 - accuracy: 0.59 - ETA: 0s - loss: 0.6793 - accuracy: 0.59 - ETA: 0s - loss: 0.6785 - accuracy: 0.59 - ETA: 0s - loss: 0.6777 - accuracy: 0.59 - ETA: 0s - loss: 0.6769 - accuracy: 0.59 - ETA: 0s - loss: 0.6760 - accuracy: 0.60 - 8s 54ms/step - loss: 0.6756 - accuracy: 0.6004 - val_loss: 0.6313 - val_accuracy: 0.6121
40/40 [==============================] - ETA: 3:39 - loss: 0.6966 - accuracy: 0.37 - ETA: 1s - loss: 0.6954 - accuracy: 0.4062 - ETA: 1s - loss: 0.6942 - accuracy: 0.45 - ETA: 1s - loss: 0.6936 - accuracy: 0.47 - ETA: 1s - loss: 0.6927 - accuracy: 0.49 - ETA: 1s - loss: 0.6917 - accuracy: 0.51 - ETA: 1s - loss: 0.6908 - accuracy: 0.52 - ETA: 1s - loss: 0.6900 - accuracy: 0.53 - ETA: 1s - loss: 0.6896 - accuracy: 0.53 - ETA: 0s - loss: 0.6887 - accuracy: 0.54 - ETA: 0s - loss: 0.6877 - accuracy: 0.55 - ETA: 0s - loss: 0.6868 - accuracy: 0.55 - ETA: 0s - loss: 0.6861 - accuracy: 0.56 - ETA: 0s - loss: 0.6852 - accuracy: 0.56 - ETA: 0s - loss: 0.6843 - accuracy: 0.56 - ETA: 0s - loss: 0.6835 - accuracy: 0.57 - ETA: 0s - loss: 0.6825 - accuracy: 0.57 - ETA: 0s - loss: 0.6814 - accuracy: 0.57 - ETA: 0s - loss: 0.6803 - accuracy: 0.57 - ETA: 0s - loss: 0.6793 - accuracy: 0.58 - ETA: 0s - loss: 0.6788 - accuracy: 0.58 - ETA: 0s - loss: 0.6783 - accuracy: 0.58 - 8s 62ms/step - loss: 0.6774 - accuracy: 0.5836 - val_loss: 0.6379 - val_accuracy: 0.6121

Trial complete

Trial summary

|-Trial ID: 4edb2e330e3fb872bc78034ef098ba6d
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 96
|-units: 48
40/40 [==============================] - ETA: 3:34 - loss: 0.6916 - accuracy: 0.60 - ETA: 1s - loss: 0.6903 - accuracy: 0.6237 - ETA: 1s - loss: 0.6894 - accuracy: 0.62 - ETA: 1s - loss: 0.6886 - accuracy: 0.62 - ETA: 1s - loss: 0.6878 - accuracy: 0.62 - ETA: 0s - loss: 0.6870 - accuracy: 0.62 - ETA: 0s - loss: 0.6861 - accuracy: 0.62 - ETA: 0s - loss: 0.6851 - accuracy: 0.62 - ETA: 0s - loss: 0.6846 - accuracy: 0.62 - ETA: 0s - loss: 0.6841 - accuracy: 0.62 - ETA: 0s - loss: 0.6830 - accuracy: 0.62 - ETA: 0s - loss: 0.6820 - accuracy: 0.62 - ETA: 0s - loss: 0.6809 - accuracy: 0.62 - ETA: 0s - loss: 0.6798 - accuracy: 0.62 - ETA: 0s - loss: 0.6785 - accuracy: 0.62 - ETA: 0s - loss: 0.6775 - accuracy: 0.62 - ETA: 0s - loss: 0.6764 - accuracy: 0.62 - ETA: 0s - loss: 0.6754 - accuracy: 0.62 - ETA: 0s - loss: 0.6745 - accuracy: 0.62 - ETA: 0s - loss: 0.6737 - accuracy: 0.62 - ETA: 0s - loss: 0.6728 - accuracy: 0.62 - ETA: 0s - loss: 0.6723 - accuracy: 0.62 - 8s 58ms/step - loss: 0.6719 - accuracy: 0.6274 - val_loss: 0.6329 - val_accuracy: 0.6121
40/40 [==============================] - ETA: 3:38 - loss: 0.6921 - accuracy: 0.59 - ETA: 1s - loss: 0.6915 - accuracy: 0.5977 - ETA: 1s - loss: 0.6906 - accuracy: 0.60 - ETA: 1s - loss: 0.6895 - accuracy: 0.60 - ETA: 0s - loss: 0.6886 - accuracy: 0.61 - ETA: 0s - loss: 0.6877 - accuracy: 0.61 - ETA: 0s - loss: 0.6868 - accuracy: 0.61 - ETA: 0s - loss: 0.6860 - accuracy: 0.61 - ETA: 0s - loss: 0.6849 - accuracy: 0.61 - ETA: 0s - loss: 0.6840 - accuracy: 0.61 - ETA: 0s - loss: 0.6829 - accuracy: 0.61 - ETA: 0s - loss: 0.6816 - accuracy: 0.61 - ETA: 0s - loss: 0.6803 - accuracy: 0.61 - ETA: 0s - loss: 0.6796 - accuracy: 0.61 - ETA: 0s - loss: 0.6781 - accuracy: 0.61 - ETA: 0s - loss: 0.6767 - accuracy: 0.61 - ETA: 0s - loss: 0.6753 - accuracy: 0.61 - ETA: 0s - loss: 0.6741 - accuracy: 0.62 - ETA: 0s - loss: 0.6729 - accuracy: 0.62 - ETA: 0s - loss: 0.6718 - accuracy: 0.62 - ETA: 0s - loss: 0.6708 - accuracy: 0.62 - 8s 56ms/step - loss: 0.6704 - accuracy: 0.6216 - val_loss: 0.6282 - val_accuracy: 0.6145

Trial complete

Trial summary

|-Trial ID: 2ea8b74ab6daad6619212dc8d64c7a2c
|-Score: 0.613296627998352
|-Best step: 0

Hyperparameters:

|-output_dim: 64
|-units: 64
40/40 [==============================] - ETA: 3:23 - loss: 0.6930 - accuracy: 0.45 - ETA: 0s - loss: 0.6922 - accuracy: 0.5122 - ETA: 0s - loss: 0.6914 - accuracy: 0.53 - ETA: 0s - loss: 0.6906 - accuracy: 0.55 - ETA: 0s - loss: 0.6899 - accuracy: 0.56 - ETA: 0s - loss: 0.6891 - accuracy: 0.57 - ETA: 0s - loss: 0.6883 - accuracy: 0.57 - ETA: 0s - loss: 0.6874 - accuracy: 0.58 - ETA: 0s - loss: 0.6865 - accuracy: 0.58 - ETA: 0s - loss: 0.6856 - accuracy: 0.58 - ETA: 0s - loss: 0.6849 - accuracy: 0.59 - ETA: 0s - loss: 0.6842 - accuracy: 0.59 - ETA: 0s - loss: 0.6833 - accuracy: 0.59 - ETA: 0s - loss: 0.6825 - accuracy: 0.59 - ETA: 0s - loss: 0.6817 - accuracy: 0.59 - ETA: 0s - loss: 0.6809 - accuracy: 0.59 - ETA: 0s - loss: 0.6799 - accuracy: 0.60 - ETA: 0s - loss: 0.6789 - accuracy: 0.60 - ETA: 0s - loss: 0.6779 - accuracy: 0.60 - ETA: 0s - loss: 0.6770 - accuracy: 0.60 - 7s 47ms/step - loss: 0.6761 - accuracy: 0.6049 - val_loss: 0.6458 - val_accuracy: 0.6121
40/40 [==============================] - ETA: 3:39 - loss: 0.6924 - accuracy: 0.59 - ETA: 0s - loss: 0.6913 - accuracy: 0.6181 - ETA: 0s - loss: 0.6907 - accuracy: 0.62 - ETA: 0s - loss: 0.6899 - accuracy: 0.62 - ETA: 0s - loss: 0.6892 - accuracy: 0.62 - ETA: 0s - loss: 0.6886 - accuracy: 0.62 - ETA: 0s - loss: 0.6883 - accuracy: 0.62 - ETA: 0s - loss: 0.6876 - accuracy: 0.62 - ETA: 0s - loss: 0.6873 - accuracy: 0.62 - ETA: 0s - loss: 0.6865 - accuracy: 0.62 - ETA: 0s - loss: 0.6855 - accuracy: 0.62 - ETA: 0s - loss: 0.6843 - accuracy: 0.62 - ETA: 0s - loss: 0.6833 - accuracy: 0.62 - ETA: 0s - loss: 0.6824 - accuracy: 0.62 - ETA: 0s - loss: 0.6815 - accuracy: 0.62 - ETA: 0s - loss: 0.6805 - accuracy: 0.62 - ETA: 0s - loss: 0.6797 - accuracy: 0.62 - ETA: 0s - loss: 0.6789 - accuracy: 0.62 - ETA: 0s - loss: 0.6781 - accuracy: 0.62 - ETA: 0s - loss: 0.6775 - accuracy: 0.62 - ETA: 0s - loss: 0.6767 - accuracy: 0.62 - 8s 55ms/step - loss: 0.6760 - accuracy: 0.6260 - val_loss: 0.6472 - val_accuracy: 0.6121

Trial complete

Trial summary

|-Trial ID: c74cc27267894ca7b096717d75c42a83
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 32
|-units: 64
40/40 [==============================] - ETA: 4:08 - loss: 0.6960 - accuracy: 0.40 - ETA: 0s - loss: 0.6946 - accuracy: 0.4614 - ETA: 0s - loss: 0.6937 - accuracy: 0.50 - ETA: 0s - loss: 0.6928 - accuracy: 0.52 - ETA: 0s - loss: 0.6919 - accuracy: 0.54 - ETA: 0s - loss: 0.6914 - accuracy: 0.54 - ETA: 0s - loss: 0.6906 - accuracy: 0.55 - ETA: 0s - loss: 0.6901 - accuracy: 0.56 - ETA: 0s - loss: 0.6896 - accuracy: 0.56 - ETA: 0s - loss: 0.6892 - accuracy: 0.57 - ETA: 0s - loss: 0.6887 - accuracy: 0.57 - ETA: 0s - loss: 0.6880 - accuracy: 0.57 - ETA: 0s - loss: 0.6876 - accuracy: 0.57 - ETA: 0s - loss: 0.6868 - accuracy: 0.58 - ETA: 0s - loss: 0.6863 - accuracy: 0.58 - ETA: 0s - loss: 0.6857 - accuracy: 0.58 - 8s 47ms/step - loss: 0.6852 - accuracy: 0.5876 - val_loss: 0.6572 - val_accuracy: 0.6121
40/40 [==============================] - ETA: 3:50 - loss: 0.6916 - accuracy: 0.62 - ETA: 0s - loss: 0.6913 - accuracy: 0.6061 - ETA: 0s - loss: 0.6905 - accuracy: 0.60 - ETA: 0s - loss: 0.6897 - accuracy: 0.60 - ETA: 0s - loss: 0.6889 - accuracy: 0.61 - ETA: 0s - loss: 0.6881 - accuracy: 0.61 - ETA: 0s - loss: 0.6872 - accuracy: 0.61 - ETA: 0s - loss: 0.6862 - accuracy: 0.61 - ETA: 0s - loss: 0.6856 - accuracy: 0.61 - ETA: 0s - loss: 0.6847 - accuracy: 0.61 - ETA: 0s - loss: 0.6838 - accuracy: 0.62 - ETA: 0s - loss: 0.6833 - accuracy: 0.62 - ETA: 0s - loss: 0.6827 - accuracy: 0.62 - ETA: 0s - loss: 0.6822 - accuracy: 0.62 - ETA: 0s - loss: 0.6813 - accuracy: 0.62 - 8s 41ms/step - loss: 0.6807 - accuracy: 0.6218 - val_loss: 0.6519 - val_accuracy: 0.6121

Trial complete

Trial summary

|-Trial ID: fabf0c5b243977705a585be11abbd890
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 32
|-units: 16
INFO:tensorflow:Oracle triggered exit
CPU times: user 3min 26s, sys: 5.52 s, total: 3min 32s
Wall time: 2min 53s

## Retrieve the best models from the tuner
models = tuner.get_best_models(num_models=2)

plot_model(models[0], show_shapes=True)
../_images/dl-simple-case_112_0.png 
## Retrieve the summary of results from the tuner
tuner.results_summary()

Results summary

|-Results in my_dir/untitled_project
|-Showing 10 best trials
|-Objective(name='val_accuracy', direction='max')

Trial summary

|-Trial ID: ae64d7ed05ffee5ffebb5db7ff7da094
|-Score: 0.6168371438980103
|-Best step: 0

Hyperparameters:

|-output_dim: 128
|-units: 48

Trial summary

|-Trial ID: 2ea8b74ab6daad6619212dc8d64c7a2c
|-Score: 0.613296627998352
|-Best step: 0

Hyperparameters:

|-output_dim: 64
|-units: 64

Trial summary

|-Trial ID: 78c668676dde2b55ddda6af9c183580c
|-Score: 0.6129032373428345
|-Best step: 0

Hyperparameters:

|-output_dim: 96
|-units: 32

Trial summary

|-Trial ID: 193383d74737cdad5c526eba4cf2dc03
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 64
|-units: 32

Trial summary

|-Trial ID: f169fc59e0f12c5e0dad3a1304f88e86
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 64
|-units: 48

Trial summary

|-Trial ID: 0c3a82a52f5438822e0f94f232c7203e
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 32
|-units: 32

Trial summary

|-Trial ID: 3cf4f74b266185196862938ef406c47a
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 32
|-units: 48

Trial summary

|-Trial ID: 4edb2e330e3fb872bc78034ef098ba6d
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 96
|-units: 48

Trial summary

|-Trial ID: c74cc27267894ca7b096717d75c42a83
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 32
|-units: 64

Trial summary

|-Trial ID: fabf0c5b243977705a585be11abbd890
|-Score: 0.6121164560317993
|-Best step: 0

Hyperparameters:

|-output_dim: 32
|-units: 16

2.11. Explanation

Train Model with the Tuned Hyperparameters

EMBEDDING_DIM = 128
HIDDEN_STATE= 48
model6 = Sequential()
model6.add(
    layers.Embedding(input_dim=vocab_size,
                     output_dim=EMBEDDING_DIM,
                     input_length=max_len,
                     mask_zero=True))
model6.add(
    layers.Bidirectional(
        layers.LSTM(HIDDEN_STATE,
                    activation="relu",
                    name="lstm_layer",
                    dropout=0.2,
                    recurrent_dropout=0.5)))
model6.add(layers.Dense(1, activation="sigmoid", name="output"))

model6.compile(loss=keras.losses.BinaryCrossentropy(),
               optimizer=keras.optimizers.Adam(lr=0.001),
               metrics=["accuracy"])
plot_model(model6)
../_images/dl-simple-case_116_0.png 
history6 = model6.fit(X_train,
                      y_train,
                      batch_size=BATCH_SIZE,
                      epochs=EPOCHS,
                      verbose=2,
                      validation_split=VALIDATION_SPLIT)

Epoch 1/20
40/40 - 8s - loss: 0.6515 - accuracy: 0.6237 - val_loss: 0.6236 - val_accuracy: 0.6168
Epoch 2/20
40/40 - 2s - loss: 0.5493 - accuracy: 0.7089 - val_loss: 0.5125 - val_accuracy: 0.7506
Epoch 3/20
40/40 - 2s - loss: 0.4635 - accuracy: 0.7777 - val_loss: 0.4676 - val_accuracy: 0.7742
Epoch 4/20
40/40 - 1s - loss: 0.4366 - accuracy: 0.7868 - val_loss: 0.4569 - val_accuracy: 0.7797
Epoch 5/20
40/40 - 1s - loss: 0.4307 - accuracy: 0.7887 - val_loss: 0.4488 - val_accuracy: 0.7860
Epoch 6/20
40/40 - 1s - loss: 0.4234 - accuracy: 0.7972 - val_loss: 0.4429 - val_accuracy: 0.7978
Epoch 7/20
40/40 - 1s - loss: 0.4182 - accuracy: 0.8025 - val_loss: 0.4384 - val_accuracy: 0.7868
Epoch 8/20
40/40 - 1s - loss: 0.4134 - accuracy: 0.8017 - val_loss: 0.4306 - val_accuracy: 0.8017
Epoch 9/20
40/40 - 1s - loss: 0.4071 - accuracy: 0.8072 - val_loss: 0.4299 - val_accuracy: 0.8002
Epoch 10/20
40/40 - 1s - loss: 0.4040 - accuracy: 0.8086 - val_loss: 0.4279 - val_accuracy: 0.8033
Epoch 11/20
40/40 - 2s - loss: 0.4015 - accuracy: 0.8088 - val_loss: 0.4254 - val_accuracy: 0.7978
Epoch 12/20
40/40 - 2s - loss: 0.4030 - accuracy: 0.8112 - val_loss: 0.4246 - val_accuracy: 0.8041
Epoch 13/20
40/40 - 2s - loss: 0.3964 - accuracy: 0.8151 - val_loss: 0.4221 - val_accuracy: 0.8025
Epoch 14/20
40/40 - 2s - loss: 0.3928 - accuracy: 0.8143 - val_loss: 0.4192 - val_accuracy: 0.8025
Epoch 15/20
40/40 - 2s - loss: 0.3901 - accuracy: 0.8159 - val_loss: 0.4203 - val_accuracy: 0.7978
Epoch 16/20
40/40 - 2s - loss: 0.3852 - accuracy: 0.8196 - val_loss: 0.4146 - val_accuracy: 0.8049
Epoch 17/20
40/40 - 2s - loss: 0.3851 - accuracy: 0.8192 - val_loss: 0.4155 - val_accuracy: 0.8049
Epoch 18/20
40/40 - 1s - loss: 0.3825 - accuracy: 0.8210 - val_loss: 0.4175 - val_accuracy: 0.8002
Epoch 19/20
40/40 - 1s - loss: 0.3834 - accuracy: 0.8145 - val_loss: 0.4179 - val_accuracy: 0.8025
Epoch 20/20
40/40 - 1s - loss: 0.3785 - accuracy: 0.8222 - val_loss: 0.4200 - val_accuracy: 0.7970

plot2(history6)
../_images/dl-simple-case_118_0.png 
explainer = LimeTextExplainer(class_names=['male'], char_level=True)

def model_predict_pipeline(text):
    _seq = tokenizer.texts_to_sequences(text)
    _seq_pad = keras.preprocessing.sequence.pad_sequences(_seq, maxlen=max_len)
    #return np.array([[float(1-x), float(x)] for x in model.predict(np.array(_seq_pad))])
    return model6.predict(np.array(_seq_pad))

text_id = 12
print(X_test_texts[text_id])
model_predict_pipeline([X_test_texts[text_id]])

Hymie

array([[0.33566964]], dtype=float32)

exp = explainer.explain_instance(X_test_texts[text_id],
                                 model_predict_pipeline,
                                 num_features=100,
                                 top_labels=1)

exp.show_in_notebook(text=True)

y_test[text_id]

1

exp = explainer.explain_instance('Tim',
                                 model_predict_pipeline,
                                 num_features=100,
                                 top_labels=1)
exp.show_in_notebook(text=True)

exp = explainer.explain_instance('Michaelis',
                                 model_predict_pipeline,
                                 num_features=100,
                                 top_labels=1)
exp.show_in_notebook(text=True)

exp = explainer.explain_instance('Sidney',
                                 model_predict_pipeline,
                                 num_features=100,
                                 top_labels=1)
exp.show_in_notebook(text=True)

exp = explainer.explain_instance('Timber',
                                 model_predict_pipeline,
                                 num_features=100,
                                 top_labels=1)
exp.show_in_notebook(text=True)

exp = explainer.explain_instance('Alvin',
                                 model_predict_pipeline,
                                 num_features=100,
                                 top_labels=1)
exp.show_in_notebook(text=True)

2.12. References

  • Chollet (2017), Ch 3 and Ch 4

1. Neural Network From Scratch 3. Deep Learning: Sentiment Analysis